Développement d'une sonde portable d'imagerie Raman pour guider l'exérèse de tumeur

One of the first line of defense against solid cancer is the surgical resection of all the cancerous tissue by a surgeon. Residual cancer tissue and cells left after the surgery directly impact patient outcome with a higher probability of recurrence, therefore complete resection is important to ensure the best patient prognosis. From the actual lack of surgical guidance instruments, most of the time the surgeon must rely on visual inspection and palpation of the tissue. Those techniques are highly subjective and their lack of sensibility for detecting cancerous cell infiltration often leads to incomplete or overly aggressive resection. There is currently a need for surgical guidance instruments that could help the surgeon perform a more complete cancer resection while improving the safety of the procedure. Raman spectroscopy has shown great promise in the development of fiber-optic probes that have the potential to help the surgeon differentiate between cancerous and normal tissue based on molecular contrast. Up to now, most development led to single point probe or needle with interrogation volumes of less than 500 μm diameter. However, spatial registration of multiple spectra through a mesoscopic wide-field Raman probe could provide clinicians with molecular contextual information to facilitate the localization of tumor margins and infiltration. Here we present two proofs of concepts towards the development of a wide-field Raman probe that could potentially help surgeons in tumor margin assessment. A first proof-of-concept study shows that spontaneous Raman spectroscopy could be achieved on a large field-of-view through a flexible cable using a coherent bundle of fibers. Reconstruction of hyperspectral images was done using a liquid crystal tunable filter combined with an EMCCD camera. We demonstrated that the developed system is able to recreate a Raman image over a field of view of 25 mm2 with a spatial resolution of 55.7 μm through a flexible cable. Each pixel of the Raman image formed a Raman spectrum covering the spectral range from 453 to 1749 cm-1 with an average spectral resolution of 95 cm-1. Individual fat and adipose tissue of porcine meat spectra of the Raman image correlated well with spectra acquired with an established single point Raman probe with an average Pearson correlation factor of 0.82. Tissue maps were reconstructed based on the molecular contrast generated by Raman spectroscopy using neural network classification. However, this first implementation was limited by the very low signal to noise ratio and a poor spectral resolution. This does not allow the detection of important spectral features required for classification of tissue with subtler molecular differences as cancerous tissue. In addition, the system implementation did not completely reflect the intention to develop a handheld probe with two separate systems to perform the illumination and detection. Nonetheless, this proof of concept was the first in the literature to show a wide-field Raman spectroscopy over a few millimeters through a coherent bundle of fibers on biological samples. A second proof-of-concept was designed to overcome the limitations of the previous system and to reach suitable specifications for clinical translation. The principal goal of this second prototype was to collect spectra comparable to the established single point Raman probe over a large field of view of a few millimeters. The same coherent bundle fiber was used, with a new detection system involving a rotating mirror combined with an imaging spectrometer to increase spectral resolution and signal to noise ratio. The probe was redesigned to combine the illumination and detection through one end. The system provided Raman spectrum with spectral resolution of 6 cm-1 in each pixel of a 14mm2 image with a spatial resolution below 115 μm. The improved spectral resolution has led to the identification of several peaks of proteins and lipids already reported in the literature over samples of calf brain and swine tissue. Individual spectra from Raman images compared well with the single point Raman probe measurements with a Pearson correlation factor above 0.85. Calf brain and porcine meat tissue maps were reconstructed using peak ratios associated with protein and lipids. Those maps clearly depicted the contrast between adipose and muscle tissue of porcine tissue and contrast between gray and white matter of calf brain. The system presented here have the potential to be translated for clinical use as an intraoperative probe or endoscope to guide the surgeon to minimize the amount of residual pathological tissue after surgery.